affectively-ai/roberta-base-go-emotions-onnx

RoBERTa GoEmotions (ONNX, int8 Quantized)

Production-ready ONNX conversion of SamLowe/roberta-base-go_emotions for in-browser emotion detection — zero server cost, zero latency, complete privacy.

Highlights

• 28 emotion labels — admiration, amusement, anger, annoyance, approval, caring, confusion, curiosity, desire, disappointment, disapproval, disgust, embarrassment, excitement, fear, gratitude, grief, joy, love, nervousness, optimism, pride, realization, relief, remorse, sadness, surprise, neutral
• ~124 MB quantized (int8 dynamic) — runs in any modern browser
• transformers.js compatible — drop-in pipeline('text-classification')
• Trained on GoEmotions — 58k Reddit comments with fine-grained emotion labels

Quick Start

import { pipeline } from '@huggingface/transformers';

const classifier = await pipeline(
  'text-classification',
  'affectively-ai/roberta-base-go-emotions-onnx',
  { dtype: 'q8' }
);

const result = await classifier('I am so happy and grateful today!', {
  top_k: 5,
});
// [
//   { label: 'joy', score: 0.95 },
//   { label: 'gratitude', score: 0.87 },
//   { label: 'optimism', score: 0.42 },
//   ...
// ]

Emotion Labels

Positive	Negative	Ambiguous
admiration, amusement, approval, caring, desire, excitement, gratitude, joy, love, optimism, pride, relief	anger, annoyance, disappointment, disapproval, disgust, embarrassment, fear, grief, nervousness, remorse, sadness	confusion, curiosity, realization, surprise, neutral

Conversion Details

Property	Value
Base model	SamLowe/roberta-base-go_emotions
Export	PyTorch → ONNX via Optimum
Quantization	int8 dynamic (ORTQuantizer, avx512_vnni)
Original size	~500 MB (fp32)
Quantized size	~124 MB

Use Cases

This model powers the emotion detection layer in Edgework.ai — bringing fast, cheap, and private inference as close to the user as possible. Ideal for:

• Real-time emotion tracking in journaling apps
• Sentiment dashboards for customer feedback
• Empathetic chatbot pre-processing
• Mental wellness check-ins

About

Published by AFFECTIVELY · Managed by @buley

We convert, quantize, and publish production-ready ONNX models for edge and in-browser inference. Every release is tested for correctness and stability before publication.

• All models · GitHub · Edgework.ai